The 10-Ads Workflow (Luke)
- Time split: 20% analysis, 60-80% ideation, 10-20% writing. Spending 40% on writing means you're doing it wrong.
- Weekly: sort account by spend, copy top ad scripts + read comments into one doc, saturate your brain (15-30 min).
- STORMING = 8 idea sources: Swipes, Templates, Organic, Research/comments, Matrix (coverage-gap grid — write to empty cells), Internal-vectors, New-styles, Gambits. Pull 1-2 per batch.
- Internal Vectors is #1: paste winners into Claude, ask for patterns, YOU read simultaneously and blend your notes with AI's.
- Organic: run a client-niche IG account; grab reels whose views wildly outlier the creator's average (~250x in KB examples). Any hook can bridge to any product.
- Battleship: spread 10 ads across buckets + sources. Once one hits, THEN make variations of that concept.
- Two awareness buckets only: 60-70% unaware/problem-aware, 30-40% solution/product-aware. Keep separate or AI blends them.
- Briefs = a few sentences, thrust only. Don't over-specify or you kill the ad's creativity. Reset the chat per ad.
Vicious Hooks (70-80% of writing time)
- Every cold hook gets a shrug ('who cares') or a shake ('oh shit, how?'). A comfortable, polite hook is dead on arrival.
- 4 ingredients of a shake: hyper-relevant to them NOW, hyper-specific, unscratchable curiosity loop, emotionally vicious stakes.
- Charged language = words that TRIGGER emotion, not describe it. 'Betrayal/grief' too nice; 'caught, affair, fart, naked, die' land.
- Germanic caveman words: sweat not perspire, bad not severe. One-two charged words per hook; stacking kills it.
- Consequence first, mechanism later. Lead with what they already care about; ions and science come after.
- Hidden-cause hook wins cold traffic: 'the symptom you blamed on X is actually caused by Y.'
- Name the belief the hook shifts in one sentence. Can't name it? The hook has none. Shift must increase desire.
- Buried-lead fix: the best hook is usually a disruptive line already mid-draft. Drag it to front, sharpen it.
- 10/10 criteria: specific+relevant, charged, high-stakes, open loop, borderline unbelievable, instantly clear. Avoid two-sentence hooks.
Creative Diversity / Andromeda
- Meta assigns each ad an entity ID fingerprint driven mostly by VISUALS. Same-looking ads = one ad to the algorithm.
- Test 5-10 different statics per concept BEFORE testing copy variations. Winning message found? Scale via more visual formats.
- Segment = Pain/Promise x Demographics x Constraints. Mix to create infinite personas; pick ones with memetic velocity.
- Changing the actor/demographic in the ad often creates a new entity ID and can 4x conversions on identical copy.
- Shrapnel effect: specificity hits adjacent readers too; vagueness hits nobody. Precise idea = broader appeal, not narrower.
- Local-maximum trap: pulling from one source optimizes one peak. Diverse sources find whole new pillars of winners.
- Aspect ratios 1:1 or 4:5 for feed. Static images can be illogical/disconnected from copy — disconnected often wins.
- Ads can test every candidate mechanism at once — map them all and run simultaneously (a VSL only ever tests one).
Copy Instincts — Pacing & Mechanism
- AI copy feels flat because it weights every sentence equally. Great copy weights: load-bearing ideas get room, filler gets breezed.
- Breathe (slow, dimensionalize): big ideas, emotion, proof, reveals. Breeze (fast): features, logistics, and usually the mechanism.
- Health VSLs spend ~10% on what/how, ~80% on what it means for the prospect's life. Weak writers invert this.
- Pick 3-5 points that MUST land. Give each its own beat, isolated line, setup and payoff. Contrast IS the emphasis.
- 4 beats of a load-bearing idea: Setup (collapse old belief via paradox) -> assert new cause -> why (a picture, not lecture) -> what it means.
- Mechanism, same order every time: name the science -> accessible metaphor -> what it fixes inside -> what it means for your day.
- One idea per ad. At ad level you get ~5% of bandwidth; extra ideas steal real estate from intensifying the winner.
- Bold/underline/caps ~1 in 3 lines for momentum. Write subheads as hooks, not labels, so skimmers still get pulled in.
- Stage, don't summarize: never 'you might regret it later' — put them IN the scene ('It's 6 months from now...'). The #1 flat-AI fix.
Big Idea / Worldview (Luke's meta-instinct)
- Don't ask 'how do I sell this?' Ask 'what must be true for this to be THE best product?' Then build (cherry-pick) that world.
- Kill the gray: any fact a competitor could also say is gray. Build differentiation from the ONE unique fact (founder, origin, odd process).
- 'Locked' test: you explain product+mechanism+proof+enemy in one breath and the listener nods before you finish.
- Magical/associative thinking runs copy, not logic: 'salmon skin regenerates, so it regenerates yours.' Stack associations (gold+cherry+Korea).
- Always hunt only/most/best/worst — superlatives make the mechanism inevitable. Cherry-pick proof, but know the narrative first.
- Word-level work: upgrade the noun. 'Trout' is ugly; 'wild-caught Korean cherry salmon' stacks premium. Pick 'peptide' over scary 'DNA.'
- Intensify to the extreme: pain cascade to death/hell, benefit cascade to heaven. Suggest it, never state it. Practice maximal to own the range.
- Cause-effect flip always generates curiosity: 'you don't age and lose collagen — you age BECAUSE you lose collagen.'
- Ride the adoption curve: aim early-majority ideas (microplastics), novel but with momentum. Match action-verb to mechanism.
AI Bot Workflow & Anti-AI Tells
- Set 3 prompts per account once: Starter (10-12 winning ads), Hook (10-12 winning hooks), Headline. Keep two versions per awareness bucket.
- Leave the Starter's 'important' slot mostly empty — AI over-weights whatever you flag. Add only genuine must-hit proof/bans.
- Mad Lib swipe for new accounts: 'keep exact structure/hook/tone, swap only the details to mine.'
- Bots won't produce vicious hooks first pass. Push: 'more vicious,' 'more emotional charge,' 'to the puking limit,' 'level three.'
- Kill AI tells: no 'not just X but Y,' no 'this means that,' no em-dash hell, no stacked triplets, no 'let's dive in,' no dead adverbs.
- Name the noun: walk every this/it/that and confirm the noun is named in the same breath. Vague pronoun = hard fail.
- Editing passes: rewrite for natural flow -> cut whole lines (not compress) -> anal fluff cut -> simplify to 3rd-5th grade -> weave in proof/intensity.
- Don't ship the sterile mean: a draft can pass every rule and still be flat. Add stance, sharpest version, concrete scene, disarm.
- Don't copy-chief test ads. Only deep-edit proven winners you're scaling. Tokens are pennies making hundreds — don't be cheap.
Review Checklist (coaches' top kills)
- Weak/non-vicious hook is the #1 diagnosis across every call. Opens with writer's voice or a category question = kill.
- Hook must pay off the lead's first 3 lines. Bait-and-switch (hook promises one frame, body delivers another) = kill.
- One unmistakable promise, stated plainly near the top. Weak verbs ('support, help, may improve') = no promise.
- Bombard with proof and move it UP: 'featured in' + study + testimonial + before/after + numbers, stacked not spread.
- One mechanism, one metaphor, named. Combination mechanisms and un-named mechanisms both fail.
- Multiple ideas smuggled into one ad = period's #1 mistake. Commit to one; the competing angle becomes its own ad.
- Curiosity over pain: pain is the pasta (boring alone), curiosity is the sauce. Flip curiosity first in the hook.
- Logic-gap pass: read every claim, mark where a skeptic says 'why does that follow?' Close with proof/mechanism or delete.
- Delete the first three words. Start at the first vicious image, not 'Hi I'm...' or 'Have you ever...'
⚡ When Stuck
- Hook feels flat -> read only your hook: do you shrug or shake? If shrug, hunt the buried lead already in your draft.
- Can't find an angle -> paste all winners into Claude for internal vectors, read them yourself, blend both lists of patterns.
- Ad isn't differentiated -> find the ONE unique fact and ask 'what must be true about the world for this to be the best product?'
- AI copy sounds robotic -> it weighted everything equally; pick the 3-5 must-land points and give each its own isolated beat.
- Not enough winners -> you're clustered on the board; spread 10 ads across all buckets and STORMING sources.
- Mechanism won't land -> run the 4 beats: name science, accessible metaphor, what it fixes, what it means for their day.
- Copy too long? -> it's too generic, not too long. Cut lines that add no new claim/proof/emotion; make it denser.